Faithful Vision-Language Interpretation via Concept Bottleneck Models
Authors: Songning Lai, Lijie Hu, Junxiao Wang, Laure Berti-Equille, Di Wang
ICLR 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Our extensive experiments on four benchmark datasets using Label-free CBM model architectures demonstrate that our FVLC outperforms other baselines regarding stability against input and concept set perturbations. Our approach incurs minimal accuracy degradation compared to the vanilla CBM, making it a promising solution for reliable and faithful model interpretation.6 EXPERIMENTS |
| Researcher Affiliation | Academia | 1Provable Responsible AI and Data Analytics (PRADA) Lab 2King Abdullah University of Science and Technology 3IRD Institut de Recherche pour le D eveloppement, Montpellier, France 4SDAIA-KAUST AI 5Shandong University |
| Pseudocode | Yes | Algorithm 1 Faithful Vision-Language Concept |
| Open Source Code | No | The paper does not provide an explicit statement about releasing the source code for the methodology described, nor does it provide a direct link to a code repository. |
| Open Datasets | Yes | We conducted a comprehensive evaluation of our approach by training our model on four diverse datasets: CIFAR-10, CIFAR-100 (Krizhevsky et al., 2009), CUB (Wah et al., 2011), and Places365 (Zhou et al., 2017). |
| Dataset Splits | No | The paper mentions using a 'training dataset' and references standard datasets (CIFAR-10, CIFAR-100, CUB, Places365), but does not explicitly provide specific train/validation/test dataset splits (e.g., percentages, sample counts, or a detailed splitting methodology) needed for reproduction. |
| Hardware Specification | No | The paper mentions model architectures (CLIP RN50, ResNet-18, ResNet-50) but does not provide specific details about the hardware (e.g., GPU/CPU models, memory amounts, or detailed computer specifications) used for running its experiments. |
| Software Dependencies | No | The paper mentions using tools like GPT-3 and CLIP, but it does not provide specific version numbers for these or any other ancillary software components needed for replication. |
| Experiment Setup | Yes | To give the details of our experimental setup, we provided Table 5, which lists the key parameters we used during our training and evaluation process. The values of these parameters have been selected based on previous research and experimental experience, and have been carefully adjusted to achieve optimal performance. Note that these parameters include not only model architecture and optimizer type, but also important settings such as learning rate, batch size, number of training iterations, etc.Table 5: Model parameter configuration. |